Hindawi Publishing Corporation
EURASIP Journal on Image and Video Processing
Volume 2009, Article ID 793756, 17 pages
doi:10.1155/2009/793756
Research Article
Multispectral Acquisition of Large-Sized Pictorial Surfaces
Anna Paviotti,1 Filippo Ratti,1, 2 Luca Poletto,1, 2 and Guido Maria Cortelazzo1
1 Department
2 CNR-INFM
of Information Engineering, University of Padova, Via Gradenigo 6/B, 35131 Padova, Italy
Laboratory for UV and X-Ray Optical Research, Via Gradenigo 6/B, 35131 Padova, Italy
Correspondence should be addressed to Anna Paviotti, [email protected]
Received 1 February 2009; Accepted 15 September 2009
Recommended by Anna Tonazzini
Multispectral acquisition of artworks has recently received considerable attention in the image processing community. Quite
understandably, so far this attention has mainly focused on paintings, given their predominant role in museum collections.
It is worth pointing out that the instrumentation and procedures used for acquiring regular paintings are not suited for the
multispectral acquisition of large-sized painted surfaces such as frescoed halls and great paintings. Given the relevance of such
artifacts, and their widespread presence in churches or historical buildings due to their social function, the problem of finding
suitable techniques for their acquisition is certainly worth addressing. This paper focuses on multispectral acquisition of largesized pictorial surfaces, systematically addressing the practical issues related to the acquisition equipment and procedure. Given
the crucial role played by the illumination in this application, special attention is given to this issue. The proposed approach is
supported by experimental results.
Copyright © 2009 Anna Paviotti et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. Introduction
In the recent years, the acquisition of multispectral images
of paintings has become quite popular in the imaging
community dealing with cultural heritage applications [1].
A first reason is that acquiring the spectral reflectance
is the most reliable asset for faithful color reproduction,
where faithfulness is measured in terms of independence
from illumination and acquisition devices. Indeed, (R, G, B)
representation depends both on the acquisition device and
the environmental conditions [2]. Independence from the
former can be achieved by means of device-independent
color spaces but, in any case, the acquisition will be
influenced by the environment, mainly by lighting [3]. For
this reason, a number of cultural heritage institutions have
carried out the construction of “digital museums” [3], that
is, the digitization of their collections, using multispectral
sensors. A second advantage of multispectral measurements
is that they permit monitoring of the conservation status
of paintings. Indeed, multispectral color acquisition is
objective, hence repeatable and furthermore, digital images
are virtually eternal, since they do not degrade over time. A
third important consideration is that multispectral imaging
allows material identification [4]. Moreover, the nonvisible
parts of the spectrum can render valuable service both in
the noninvasive diagnosis of the conservation status of a
work of art and in the reconstruction of its history. For
example, infrared (IR) reflectography has been used for
some decades for the detection of underdrawings, while
ultraviolet (UV) fluorescence highlights former restoration
interventions. Spectral signatures of color pigments used in
the painting process can also be found in the nonvisible
spectrum. Unobtrusive diagnosis can therefore lead to virtual
restoration planning. The VASARI, MARC, and CRISATEL
projects were pioneering projects funded by the European
Commission which made use of multispectral data for the
acquisition and monitoring of paintings [5–22].
VASARI and MARC considered the multispectral acquisition of paintings in order to derive reliable CIELAB
coordinates [7, 10]. A remarkable multispectral imaging
project involving the CRISATEL scanner is the “Mona Lisa
project”. In 2004, an unprecedented number of techniques
were applied to the analysis of the Mona Lisa [23]. The
methodologies adopted included radiography, X-ray fluorescence, Raman spectrometry, digital photography under
raking light, infrared reflectography, multispectral imaging,
and 3D reconstruction. The study was aimed at assessing
and recording the conservation status of the painting and
2
at gaining an insight into Leonardo’s pictorial technique.
Multispectral digitization was performed by the French
Center of Museum Research with the camera developed
within the CRISATEL project [24]. This multispectral camera comprises 13 filters with a bandwidth of 40 nm spanning the spectrum from 380 nm to 1050 nm. Multispectral
measurements were used to identify the various colors of
the painting through luminance and color segmentation, to
relight the painting with different illuminants and to identify
restorations from the most superficial layers (near UV) to the
deepest ones (near IR).
Several techniques are currently employed to perform
noninvasive spectroscopy in cultural heritage applications.
Three of the most common are Fiber Optics Reflectance
Spectroscopy (FORS), Imaging Spectroscopy (IS) [25], and
Narrow Band Illumination (NBI) spectroscopy. The first
approach uses fiber optics to convey light (typically exiting a
monochromator) to the target object. Reflectance information is recovered in two possible ways: when a dual-beam
approach is used, the incident light beam is split before being
directed onto the object, and the unreflected beam is used
as a reference to calculate the reflection coefficient. In the
single-beam approach, the absolute reflected light intensity
is measured. In this case, reflectance can be calculated after
acquiring a reference white object. Dual-beam fiber optics
spectrometers are very accurate but not easily portable.
On the contrary, single-beam fiber optics spectrometers are
more compact and easily transportable but can lead to
greater measurement errors.
FORS instrumentation operates in a pointwise fashion.
On the contrary, IS determines spectral reflectance data for
each pixel in a spatial image. Imaging spectrometers can be
divided into two classes according to the way multispectral
information is recovered. A first approach consists in putting
a set of filters in front of the detector [5, 24, 26–28].
The number of filters used in state-of-the-art solutions
varies from 7 to 32, and the spectral resolution from
10 nm (narrow-band filters) to 40 nm (wide-band filters)
[1]. Another approach consists in using a dispersive element
to separate the different light components, which are then
detected by a sensor (typically a CCD). In this case, higher
spectral resolution (<1 nm) can be achieved, but the received
signal intensity is considerably lower than in the filter-based
approach.
Imaging spectroscopy allows the recovery of the spectral
reflectance of the whole target object (or parts of it).
Therefore, not only can a faithful color reproduction of the
object be achieved, but also more interesting tasks, such as
the comparison of the spectral content of different parts of
the object, or image segmentation based on spectral features,
can be performed. IS methods are generally less spectrally
accurate than FORS ones, but allow a more exhaustive
inspection of the spectral content of the target. It could
be claimed that IS methods are sufficiently accurate if they
allow the detection of significant spectral features in localized
parts of an object, to which more accurate analysis can be
restricted. In this logic, IS and FORS can be considered
as complementary. The reliability of the spectral content
analysis clearly depends on the accuracy of the spectral
EURASIP Journal on Image and Video Processing
reflectance reconstruction, which in turn is determined by
several factors, such as the instrument’s spectral resolution,
the measurement noise, and the choice of the illumination
source.
NBI spectroscopy is similar to IS, with the difference that
wavelength selection is performed directly at the illumination source by using LEDs, or by placing a monochromator
or a set of filters in front of a broadband source. The
use of NBI avoids the degradation of the image resolution
due to the use of filters in the optical path between the
target and the camera in filter-based IS [29]. However,
this is not an issue when using a dispersive element to
separate wavelength components. NBI spectroscopy also has
the advantage of limiting the exposure of the target surface
to light, which may be desirable for conservation purposes.
As for spectral resolution, the use of LEDs and filters
presents the same disadvantages as filter-based IS, while
higher spectral resolution (<10 nm) can be obtained by using
a monochromator. However, the use of a monochromator
may be impractical when deploying extended illumination
sources to illuminate large surfaces.
All throughout history, frescoed halls or wall-sized paintings have represented a relevant aspect of artistic expression
together with regular-sized paintings performed on wood or
canvas. Their large dimensions have made them particularly
suitable to effectively convey political, religious, or social
messages, a function that is similar to that of contemporary
large billboards or movie screens. Therefore, size has always
been considered as an important means of communication
by artists, historians, restorers, and public.
An important and often underestimated consideration
regarding the multispectral acquisition of large-sized painted
surfaces is that performing it using the instruments currently
employed for the multispectral acquisition of paintings [5,
24, 30, 31] is not an optimal choice. First of all, in those
instruments the camera is placed close to the painting
surface, and a full image is acquired moving the camera by
means of linear translators. This solution is unpractical when
acquiring surfaces of several squared meters; whereas the use
of a rotating stage allows more compactness. Most of all,
state-of-the-art instruments for the multispectral scanning
of paintings feature illumination sources positioned near the
camera so as to illuminate the small area under consideration, generally using halogen lamps due to their broad
emission spectrum [5, 30, 31]. In the case of the acquisition
of large scenes, one needs the illumination sources not to
be integral with the scanner, but of standalone type and
positioned several meters far from the scanned surface.
This particular setup forces nonstandard choices for the
illumination sources. As we will see, the emission of broad
spectrum lamps such as halogen and incandescence lamps
has a drop in the blue region. While this may not be a
problem when the lamps are positioned very close to the
target, it turns out that when they are far from the acquired
surface the signal in the blue region possibly falls close to or
below the dark signal (we recall that illumination intensity is
inversely proportional to the square of distance).
In this paper, we take into consideration fundamental
issues related to the multispectral acquisition of large-sized
EURASIP Journal on Image and Video Processing
3
Matrix detector
Imspector optics
Spatial axis
Spectral axis
Target
Objective lens
Entrance slit
Figure 1: The SPECIM Imspector V10 spectrograph [35].
surfaces such as those found in frescoed halls or large paintings. The required characteristics of the acquisition equipment and a suitable measurement procedure are defined.
Given the critical role of illumination, especially as concerns
the use of distant light sources, particular attention is devoted
to illumination issues. The considered illumination setups
are evaluated according to the corresponding reflectance
reconstruction results. In the literature, the latter are usually evaluated by showing qualitative spectral reflectance
reconstruction results (e.g., a figure comparing the reference
and measured spectral reflectance), and presenting quantitative results concerning the reference and reconstructed
(L∗ , a∗ , b∗ ) or (X, Y , Z) coordinates [30–33]. An important
contribution of this work is the proposal of a performance
characterization metric based on spectrally dependent errors
and error uncertainties, which allows a local evaluation
of the reconstruction performance along the considered
spectral window. The error average and standard deviation
are derived from those of the input signals through error
propagation, in conformity with the uncertainty evaluation
procedures recommended by metrological standards [34]. A
review of the related literature will be done in Section 3.1.
The paper is organized as follows. Section 2 introduces
the imaging tools needed for the acquisition of large-sized
painted surfaces and a suitable measurement procedure.
Section 3 describes the adopted spectrograph performance
characterization procedure. Section 4 relates on the application of this procedure to the choice of the illumination
sources. Section 5 presents the validation of two illumination
setups in laboratory and on-field acquisitions of cultural
heritage artifacts. Lastly, Section 6 draws the conclusions.
2. Imaging Equipment and Acquisition
Procedure for Large-Sized Pictorial Surfaces
In order to acquire large-sized surfaces, the imaging equipment will feature two main ingredients, namely, a multispectral IS acquisition device, which can be considered as the
core of the instrument, and a rotation mechanism for easily
covering large surface areas.
As for the multispectral camera, our choice has fallen
on the imaging spectrograph Imspector V10 by SPECIM.
ImSpector is a direct sight imaging spectrograph provided
with a dispersive element that can be quickly combined with
a broad range of industrial and scientific monochrome area
cameras to form a spectral camera. Compared to conventional color cameras and other filter-based imaging systems,
ImSpector produces full contiguous spectral information
with high quality spectral and spatial resolution. It can
cover a broad spectral range over which it enables flexible
wavelength selections via software.
A schematic drawing of the imaging spectrograph is
shown in Figure 1. The objective lens focuses the image of
the target to be acquired on the plane of the input slit of the
spectrograph. The light coming from a rectangular narrow
strip conjugated with the slit enters into the spectrograph
is dispersed by a dispersive element and focused on the
plane of the 2D detector. With reference to Figure 1, the
horizontal axis of the camera is the spatial axis, while the
vertical axis is the spectral axis. The light coming from the
same point on the target but with different wavelengths is
focused on the sensor on the same column (i.e., the same
spatial position) but in different rows (i.e., different spectral
positions). A 2D spectral image can be recovered using a
series of monochromatic images of a 2D region on the target
obtained by scanning it in the direction perpendicular to
the slit [36]. The choice of the objective focal length is
imposed by the required field-of-view (FOV) in the direction
parallel to the slit. Several objectives are currently available.
We normally use a Canon f = 16 mm, f /# 1.4, that matches the f /number of the spectrograph, or a Canon f =
25 mm, f /# 1.4 objective. The nominal spectral window of
the spectrograph is 400 to 1000 nm. The entrance slit size
is 9.8 mm × 25 μm. The sensor is the Hamamatsu C848405G. It is a progressive scan interline CCD with microlenses,
1024 (spatial) × 1344 (spectral) pixel format, 6.45 μm ×
6.45 μm pixel size, 6.6 mm × 8.7 mm active area. The CCD is
moderately cooled to reduce the thermal noise and increase
the sensitivity. The camera has 10 e− rms (effective value of
alternating voltage, expressed in electrons) readout noise,
4
dynamic range of 1800 : 1, and quantum efficiency in
the 0.35–0.70 range in the 400–750 nm spectral interval. Its
performances are significantly higher than a conventional TV
camera. The FOV in the direction parallel to the slit is limited
by the number of pixels in the spatial direction. A spatial
sampling of 2 mm gives a FOV of about 2 m at a distance of
about 5 m with the f = 16 mm objective.
Several solutions are possible to move the spectrograph
in the direction perpendicular to the slit, so as to acquire
a full 2D image. As previously mentioned, the one that
allows greater compactness when scanning large surfaces is
a rotating stage. In our setup, we have chosen to use the
Physik Instrumente M-062. The scanning in the direction
perpendicular to the slit is performed at constant angular
steps. This means that the spatial pixel resolution increases
with the incident angle between surface normal and scanning
direction. However, in practice the maximum incident angle
does never get large enough for these variations to be
significant. For example, working at a distance of 5 m, with
an angular step of 0.5 mrad, and performing 500 acquisitions
symmetrically with respect to the surface normal (these were
the acquisition parameters used to acquire the fresco in
the Castello del Buonconsiglio; see Section 5.2), we reach a
maximum incident angle of 125 mrad. This means the spatial
pixel resolution varies from 2.5 to 2.54 mm.
The variation of the incident angle also influences the
acquired spectral reflectance, as the amount of light reflected
by a surface depends both on the lighting and sensing
directions [37]. We implicitly assume that all the acquired
surfaces are approximately Lambertian, and thus that the
reflectance does not vary with the observation direction.
From a practical point of view, this means we should be
careful when acquiring, for example, frescoes with extended
gold leaf coverage. As for possible reflections from the
surrounding surfaces, we assume these are negligible in
comparison with the light from the illumination sources.
What is not accounted for is the dependence of the spectral
reflectance on the incident light direction. Therefore, in
order to compare different acquisitions for monitoring
purposes, the operator should place the illumination sources
in matching positions.
A suitable acquisition procedure can be devised remembering that our imaging spectrograph measures absolute
light intensities. Therefore, in order to recover spectral
reflectance it is necessary to acquire a reference white signal
under the same illumination conditions as the target object.
This is generally performed by using a white screen of known
reflectance superimposed to the target surface. While this
is a straightforward task when acquiring a regular painting,
several challenges arise when acquiring a larger surface.
First of all, the dimensions of a screen covering the whole
surface can become prohibitive. Moreover, some parts of the
surface, consider, for example, the topmost parts of a wall
or of ceilings, may not be reachable. A possible solution
is therefore that of acquiring the white signal in a piecewise manner as far as the target surface can be reached,
performing some kind of interpolation during the processing
step. A moving support for the screen must therefore be
available. It is also worth pointing out that if more complex
EURASIP Journal on Image and Video Processing
surfaces are scanned, for example, frescoed vaults, the subtle
changes in illumination may not be accurately measurable at
all.
The spectral reflectance of the target for any pixel Rn,m
within the sensor is calculated by
Rn,m = η
Sn,m − sm
,
Wn,m − wm
(1)
where m = 1, . . . , M (M = 1024) spans the spatial dimension
and n = 1, . . . , N (N = 1344) spans the spectral dimension,
Sn,m is the color signal (the raw signal collected by the
sensor), Wn,m is the reference white signal and η is the
reference white albedo (the albedo is defined as the spectral
reflectance of a Lambertian surface). The symbols sm and wm
represent the signal-dependent offsets for the color signal
and the reference white, respectively. They are calculated as
the mean of the first four bands of the acquired signal (lying
in the blue region: 393–409 nm), where no useful signal is
present due to the low sensitivity of our CCD. Formalizing,
we have
1
Xn,m ,
4 n=1
4
xm =
(2)
where X ∈ {S, W }, x ∈ {s, w}. The choice of signaldependent offsets or dark signals allows to compensate for
possible nonlinearities of the camera, which cause the dark
current signal to be dependent on the useful signal. Dark
current values are always signal and pixel dependent [38].
The most correct procedure would be that of subtracting
to each pixel the corresponding dark current value [30,
31], acquired at the same temperature conditions as the
active frame. As this would require doubling the number of
acquisitions for each frame, a good compromise [38] is that
of using inactive pixels to compute a mean value for the dark
current. Averaging the offset along several samples allows to
isolate the contribution of the dark current from that of the
dark current random noise term.
As a last observation, we would like to mention that
the choice of this setup for the multispectral camera allows
to easily integrate it with other scanning instruments. For
example, our imaging spectrograph is coupled with a timeof-flight laser scanner acquiring 3D information. In practical
situations, the use of 3D data is particularly useful when
dealing with large painted areas, since they are often piecewise planar rather than planar (e.g., frescoed rooms), and
sometimes present a more complex structure (e.g., painted
vaults). The architecture of the instrument can be seen in
Figure 2. Such an instrument has been employed for the
acquisition of small frescoed architectural volumes [32, 39].
However, the illumination setup chosen at the time, featuring
two metallic iodide lamps, offered unsatisfactory results in
the reconstruction of spectral reflectance, and eventually
motivated the presented study.
2.1. Spectral Calibration. The spectral calibration of an imaging spectrograph consists in determining the correspondence
between pixel indices and wavelengths. The calibration has
EURASIP Journal on Image and Video Processing
5
Range-finder
Range-finder
Mirror for laser
vertical scanning
Laser beam
Rotating mirror
Entrance slit
Spectrophotometer
Rotating stage
Image
spectrograph
Rotating stage
for horizontal
scanning
(a)
(b)
Figure 2: Schematic (a) and picture (b) of the instrument coupling the multispectral camera with a time-of-flight laser scanner.
1100
been performed by measuring the spectra of low pressure gas
lamps which emit very narrow spectral lines characteristic
of the gas in the lamp. The correspondence between pixels
and wavelengths has been found to be nearly linear (see
Figure 3). The spectral dispersion is 72.1 nm/mm and the
spectral resolving element is 0.465 nm/pixel.
900
800
λ (nm)
3. Performance Characterization for
the Imaging Spectrograph
1000
700
600
3.1. Related Work. An extensive literature has addressed the
problem of characterizing the performance of multispectral
measurement systems. Some works [40, 41] have addressed
the problem of diagnosing and correcting systematic spectrophotometric errors. In particular, Berns and Petersen
[41] have defined a method for identifying and correcting
the effects of seven types of spectrophotometric errors on
multispectral measurements. We do not consider the effects
of spectrophotometric errors in our analysis but concentrate
on the influence of the illumination. Spectrophotometric
errors clearly affect the systematic (average) errors found
in our performance characterization procedure. However,
since all measurements have been performed using the same
instrument under the same experimental conditions, they
do not affect the comparison between different illumination
sources.
Some works have also studied the propagation of random errors to color measurements [42–44]. However, their
analysis is mainly concerned with propagation to colorimetric coordinates. Fairchild and Reniff [43] have calculated
wavelength-dependent errors on reflectance as an intermediate step to propagation to (X, Y , Z) and (L∗ , a∗ , b∗ )
coordinates. However, they consider all input quantities to
be uncorrelated, and the dark current term to have zero
uncertainty. Our analysis can be considered an extension of
this method, accounting for correlated measurement errors
and error-affected dark currents. Burns and Berns [44] have
also extended the method in [43] to account for correlated
500
400
300
0
200
400
600
800
1000
1200
1400
Pixel
Figure 3: Spectral calibration curve.
measurement errors. However, they have only considered
error propagation to CIELAB coordinates, without explicitly
considering errors on spectral reflectance. Moreover, their
analysis has been performed by considering uncorrelated
instrument errors, and only accounting for correlation
between color matching functions.
The effects of noise on multispectral measurements
under different illumination sources have been addressed by
Vrhel and Trussell [45]. Their purpose is to select the set
of transmittance filters which minimizes trichromatic color
measurement errors wrt a set of a set of fixed illuminants.
On the contrary, our aim is that of selecting the illumination
source which minimizes wavelength-dependent errors on
spectral reflectance.
3.2. Performance Analysis. As previously mentioned, we
have chosen to characterize the reflectance reconstruction
6
EURASIP Journal on Image and Video Processing
performance by the error and the error standard deviation
calculated as functions of wavelength. We have performed
five multiple independent acquisitions of a set of colored
calibrated tiles (Spectralon, LabSphere [46]), with spectral
reflectance functions specified by the manufacturer, and
compared the tabulated and measured spectral reflectance
functions. The error standard deviation (uncertainty) has
been derived from the statistics of the input signals. The
tile set comprises eight colored tiles spanning the visible
spectrum (red, orange, yellow, green, cyan, blue, violet,
and purple), with spectral reflectance tabulated in the 380–
830 nm interval. As a reference white, we have used a
Labsphere white tile with a flat 80% spectral reflectance over
the interval 250–2500 nm.
The spectral reflectance for the ith tile has been computed by considering the color and white signals averaged
over the J = 5 repeated independent acquisitions, so as to
decrease the input signal noise. Equation (1) then becomes
Rin,m
=
J
J
i, j
i, j
j =1 Sn,m − (1/J)
j =1 sm
0.8
J
i, j
i, j
(1/J) j =1 Wn,m − (1/J) Jj =1 wm
(1/J)
,
(3)
R = 0.8
(1/J)
(1/J)
i
= 0.8
J
J
i, j
j =1 S
− (1/J)
i, j
− (1/J)
j =1 W
J
i, j
j =1 s
J
j =1 w
i, j
σx ,xk =
5 1
J(J − 1) j =1
5 1
J(J − 1) j =1
σ X ,xk = σ xk ,X =
S −s
i
W −w
Ri = 0.8
Si
i , W
i .
= f S
(4)
(5)
The reflectance error for each of the eight colors can now
be defined as
i
,
ei = Ri − Rtab
j
j
j
x − x · xk − x k ,
J 1
J(J − 1) j =1
j
(7)
j
(6)
where
is the tabulated spectrum of the ith tile.
We want to calculate the variance of the error from
the variances of the experimentally observed variables
through error propagation [34]. The observed variables are
{Si, j } j =1,...,J and {Wi, j } j =1,...,J , but it will be useful to consider
also the newly defined variables si, j and wi, j . For the sake
of simplicity, we would like to apply the error propagation
X ∈ {S, W}. We thus
formula to the derived variables X,
X − X · xk − xk ,
σ 2X = σX2 + σx2 1N + 2σ X ,x ,
σ X ,X k = σX ,Xk + σx ,xk 1N + 2σ X ,xk ,
(8)
where 1N is a vector storing N unitary values.
The variance of the derived variable ei can finally be
calculated as (see [34])
σ 2ei
=
∂f
∂Si
2
σ 2Si
∂f
+
i
∂W
2
σ 2W
i
(9)
∂f ∂f
2
+2
σ i i + σRtab ,
i S ,W
∂Si ∂W
where the partial derivatives can be easily computed from (5)
as
∂f
0.8
(10)
=
,
i
i
W
∂S
∂f
0.8 · Si
= − 2 .
i
∂W
i
W
.
i
i
W
j
X − X · Xk − Xk ,
where , k ∈ {1, 2} and X1 = S, X2 = W, x1 = s, x2 = w. The
products in (7) are intended element-wise. The underlying
model to these equations is that of an observation given by
the sum of the “true” spectrum with a random noise.
we have that
From the definition of X,
i
X − x. Equation (4) then becomes
We now define X
i
Rtab
σX ,Xk =
where i = 1, . . . , 8 is the tile index and j = 1, . . . , J, J = 5, the
repetition index. We recall that the index n spans the spectral
dimension, while m spans the spatial dimension. Although
the CCD sensor acquires 1344 spectral values for each spatial
pixel, we have considered N = 112 bands averaged over 12sample bins to lower the acquisition noise. Therefore, from
now on we will consider n = 1, . . . , N, N = 112. Moreover,
in the following we shall drop the spatial index m, with the
understanding that we are considering the central pixel of
each tile. Introducing the vector notation X [Xn ]n=1,...,N ,
(3) can be rewritten as
i
W
from those
need to derive the variance and covariance of S,
of S, W, s, and w. In accordance with (4), we have to consider
the covariance of the mean vectors S, W and of the mean
variables s and w, which can be calculated as
(11)
The variable σR2tab is the variance of the tabulated reflectance.
As it is less than 2.5 · 10−5 (from the calibration certificate
of the Spectralon dataset [46]), it can be neglected in the
computation of σ 2ei .
Once we have obtained the error and error variance for
each tile, we can average them over the eight tiles to obtain
the average error (AE) and the average error variance as
1 i
e,
8 i=1
8
e=
1 2
σ i.
8 i=1 e
8
σ 2e =
(12)
However, as the standard unit of uncertainty is standard
deviation [34], we will eventually use the average error
standard deviation (AESTD) instead of the variance. The
AESTD is given by
1
σ ei ,
8 i=1
8
σe =
(13)
EURASIP Journal on Image and Video Processing
7
×103
×103
4
4
3
Counts
Counts
3
2
1
1
0
2
400
500
600
700
λ
800
900
0
380
1000
400
420
440
(a)
×103
3
3
500
520
4
Counts
4
Counts
480
(b)
×103
2
1
0
460
λ
2
1
400
500
600
700
λ
800
900
1000
(c)
0
400
500
600
700
λ
800
900
1000
(d)
Figure 4: Emission spectrum of the three lamps used in our study: (a) metallic iodide lamp, (b) metallic iodide lamp (blue region), (c)
halogen lamp, and (d) incandescence lamp.
which is close to, but does not coincide with the square root
of the variance expressed in (12).
4. Illumination Selection for Distant
Light Sources
The light reflected by an object depends both on the object’s
surface reflectance characteristics and on the composition
of the light illuminating the object. An ideal illumination
source for the measurement of an object’s spectral reflectance
should thus exhibit a uniform spectrum over the whole
desired wavelength range. Unfortunately, no existing illumination apparatus satisfies this condition over the visible
and near-infrared spectrum, so that suboptimal solutions
must be adopted. As previously mentioned, state-of-theart solutions for the multispectral acquisition of paintings
mostly employ halogen lamps due to their smooth and broad
spectrum. However, in our case, the inverse-square law for
light intensity, combined with the low sensitivity of the CCD
in the blue region and with the emission drop of halogen
lamps for those wavelengths, causes the white signal to fall
close to the dark signal in the blue region. This consideration
has suggested the use of two different lamps, one with a
broad, smooth spectrum covering the green-red region, and
one with a strong emission in the blue region.
Three different lamps have been considered in our
experiment:
(1) a Disano 250 W metallic iodide lamp (Figures 4(a)
and 4(b));
(2) a Cixi Zhongfa Lamps 500 W halogen lamp
(Figure 4(c));
(3) an Osram 100 W incandescence lamp (Figure 4(d)).
All the lamps are divergent directional sources. Figure 4
shows the emission spectrum of the lamps. These spectra
have been measured by acquiring a white target of known
reflectance illuminated by each source. The pictures represent the spectral power density of the lamps, expressed in
CCD counts.
During the acquisitions, we have set the camera exposure
time and shutter aperture so as to have a strong signal in
the blue region when using the metallic iodide lamp. This
has caused the signal to saturate in the 530–680 nm interval
(Figure 4(a)). However, as we will see, this spectral window
will be discarded for the metallic iodide lamp. A zoom of the
emission spectrum of the same lamp in the blue region can
be seen in Figure 4(b).
As shown in Figures 4(c) and 4(d), the halogen and
incandescence lamps mainly emit in the red-NIR spectrum.
8
EURASIP Journal on Image and Video Processing
The two lamps exhibit similar spectra. However, it can be
noted that the incandescence lamp has a stronger emission
in the NIR region and that its spectrum is slightly smoother.
Four illumination setups have been considered.
(1) Halogen lamp alone. This setup has been considered
for comparison with the results presented in the
literature.
(2) Incandescence lamp alone.
(3) Metallic iodide lamp and halogen lamp in a sequence.
The reflectance measurement has been obtained by
juxtaposing the spectral reflectance measured with
the metallic iodide lamp in the 420–500 nm interval
and that obtained with the halogen lamp in the 500–
800 nm interval.
(4) Metallic iodide lamp and incandescence lamp in a
sequence, obtaining the spectral reflectance as in the
previous case.
The two combined setups can be considered as a particular kind of bracketing [47], a high dynamic range imaging
technique where acquisitions taken at different exposure
times are combined to achieve better reconstruction results.
The use of two different lamps for the acquisitions (first
the saturated metallic iodide lamp, then the halogen or
incandescence lamp) is preferable both to the use of two
different exposure levels for the metallic iodide lamp, and to
a double acquisition with one of the broad spectrum lamps.
Obtaining a sufficiently strong signal in the red and blue
regions, respectively, would require very long exposure times.
As a consequence, the dark current contribution would be
increased [48], and the acquisition times would become too
long.
When using the combined setups, a scaling has been
performed in order to eliminate the discontinuity between
the reflectance acquired with the metallic iodide lamp
and that measured with the other lamp. This scaling is
always performed when measuring real reflectances, therefore the corresponding performance characterization results
are meaningful.
For each of these setups, the AE and AESTD have
been computed as described in Section 3. The performance
characterization results are discussed in the following.
Figure 5 shows the performance characterization results
for the four illumination setups described above. In Figure 6,
the AESTD is separately shown. In Figure 5(a), the halogen
lamp is considered. It can be noted that in the blue region,
the AE presents a variable behavior, that tends to get worse
the smaller the wavelength. Most of all, the AESTD greatly
increases below 480 nm (Figure 6(a)). The reason is that in
this region the illumination signal falls close to the dark
signal, so that the propagation coefficients of (10) and (11)
become larger (note that the absolute value of the coefficient
2 , where the division is intended
of (11) grows like 1/ W
component-wise).
A similar consideration can be applied when considering the AESTD of the incandescence lamp used alone
(Figure 5(b)). As shown in Figure 6(b), in this case the
AESTD is even greater (>0.1), probably due to the fact that
the emission spectrum of this lamp is slightly translated
towards the red-NIR region with respect to that of the
halogen lamp (see Figure 4). However, for λ > 500 nm the
incandescence lamp outperforms the halogen lamp, with an
AE which is nearly zero between 500 and 700 nm, and below
0.02 (in absolute value) from 700 to 800 nm.
Figure 5(c) shows the results of juxtaposing the
reflectance acquired with the metallic iodide lamp between
420 and 500 nm (the exact transition wavelength is
λ = 502 nm), and that measured with the halogen lamp
from 500 to 800 nm. In the blue region, the AE is fairly
constant and equal to about 0.01 in absolute value. As seen
in Figure 6(c), the AESTD is greatly improved (∼0.01),
showing no significant variation with wavelength. The
best results are shown in Figures 5(d) and 6(d), when the
combined advantages of the metallic iodide lamp in the blue
region and of the incandescence lamp in the green-red-NIR
regions are benefited. In this case, the AE is nearly zero
between 420 and 700 nm, and below 0.02 (in absolute value)
between 700 and 800 nm. The AE for λ = 420–500 nm
looks different in Figures 5(c) and 5(d) because of the
aforementioned scaling. Observe that the scaling has no
effect on the AESTD.
To emphasize the comparison between the use of the
incandescence and the halogen lamps, the tabulated and
reconstructed spectra for the two combined setups are
shown in Figure 7. Although both configuration lead to good
qualitative matching between true and measured spectra, the
incandescence lamp outperforms the halogen lamp in the
red-NIR region.
Summarizing, the illumination setup featuring the
metallic iodide lamp and the incandescence lamp seems the
best to be employed for the acquisition of large painted
surfaces. However, it must be considered that halogen lamps
have a number of practical advantages over incandescence
bulbs, as they are more energy-efficient, produce a higher
light intensity with the same power, and last more. For all
these reasons, in on-field applications it is more feasible
to use halogen lamps. We have therefore considered the
combination of the metallic iodide and the incandescence
lamps a sort of ideal case, which has been validated in the
acquisition of three paintings performed in a controlled
environment (see Section 5.1). On the contrary, for on-field
acquisitions we have chosen to use the metallic iodide lamp
in combination with the halogen lamp.
5. Validation in Cultural Heritage Applications
We are now going to present some results obtained by
employing two of the considered illumination setups for
the acquisition of paintings. The acquired works of art
have been chosen in order to offer a wide variety of pictorial techniques and surfaces. Section 5.1 shows the results
obtained acquiring three paintings by Italian contemporary
artists. This first set of acquisitions has been performed in
one of our laboratories, that is, in a confined, controlled
environment. It has been designed to validate our choice of
9
0.2
0.2
0.15
0.15
0.1
0.1
0.05
0.05
e ± σe
e ± σe
EURASIP Journal on Image and Video Processing
0
0
−0.05
−0.05
−0.1
−0.1
−0.15
−0.15
−0.2
450
500
550
600
650
700
750
−0.2
800
450
500
550
600
λ
0.2
0.2
0.15
0.15
0.1
0.1
0.05
0.05
0
−0.05
−0.1
−0.1
−0.15
−0.15
500
550
750
800
650
700
750
800
0
−0.05
450
700
(b)
e ± σe
e ± σe
(a)
−0.2
650
λ
600
650
700
750
800
−0.2
450
500
550
600
λ
(c)
λ
(d)
Figure 5: AE ± AESTD for (a) the halogen lamp alone, (b) the incandescence lamp alone, (c) the metallic iodide lamp and the halogen lamp
used in a sequence, and (d) the metallic iodide lamp and the incandescence lamp used in a sequence.
the illumination setup for distant light sources. Therefore,
in this particular case, the size of the acquired surfaces is
of secondary importance. Section 5.2 describes an on-field
validation of our measurement procedure, consisting of the
acquisition of a large portion of a fresco in the Castello del
Buonconsiglio in Trento (Italy).
For all the acquisitions, we have used as a reference white
a white screen of known reflectance and approximately Lambertian behavior. As multispectral measurement achieves
independence from the illumination source, an illuminant
must be chosen to display the acquisition results in the usual
trichromatic (R, G, B) coordinates. All the reconstructions
presented in this section have been obtained by choosing as illuminant the CIE standard illuminant D65 [49],
which approximates daylight illumination conditions. The
(X, Y , Z) coordinates for each measured spectral reflectance
have been computed using the CIE standard 1931 standard
observer. A conversion from (X, Y , Z) to sRGB coordinates
has finally been performed.
5.1. Acquisitions in a Controlled Environment. For the laboratory acquisitions, we have chosen the best-performing
illumination setup, consisting of the metallic iodide lamp
and the incandescence lamp used in a sequence.
The first acquired painting, “Tulips” is an oil on wood
by the contemporary Italian painter and engraver Matteo
Massagrande (Padua, 1959). The author’s interests lie in
the study of ancient pictorial techniques, of engraving and
in the art of restoration. His frequent journeys abroad
are often occasions to develop pictorial cycles and great
EURASIP Journal on Image and Video Processing
0.1
0.1
0.09
0.09
0.08
0.08
0.07
0.07
0.06
0.06
σe
σe
10
0.05
0.05
0.04
0.04
0.03
0.03
0.02
0.02
0.01
0.01
0
450
500
550
600
λ
650
700
750
0
800
450
500
550
0.1
0.1
0.09
0.09
0.08
0.08
0.07
0.07
0.06
0.06
0.05
0.04
0.03
0.03
0.02
0.02
0.01
0.01
450
500
550
600
λ
700
750
800
650
700
750
800
0.05
0.04
0
650
(b)
σe
σe
(a)
600
λ
650
700
750
800
(c)
0
450
500
550
600
λ
(d)
Figure 6: AESTD for (a) the halogen lamp alone, (b) the incandescence lamp alone, (c) the metallic iodide lamp and the halogen lamp used
in a sequence, and (d) the metallic iodide lamp and the incandescence lamp used in a sequence.
compositions. His first exhibition dates back to 1973 and
since then he has exhibited his works in several collective and
personal exhibitions, and has been awarded several prizes.
He currently lives in Padua and works between his studies
in Padua and Hajos (Hungary).
For the acquisition of “Tulips” the two lamps have been
positioned at approximately 45◦ with respect to the object
plane, so as to minimize the reflections. The (R, G, B) reconstruction of “Tulips” can be seen in Figure 8. The quality
of the reconstructed colors is very good. The resolution
is limited along the horizontal axis by the rotating stage
acquisition step.
Figure 9(b) shows the spectral reflectance of four points
indicated in Figure 9(a). Point 2 has the typical behavior
of a red, with a high reflectance over the 620–800 nm
interval and almost no light reflected in the blue region,
can be recognized. The grey points 1 and 4 have a constant
spectral reflectance over most of the visible spectrum, with
an increasing albedo from darker to lighter shades. Point 3
is a composite color with characteristics of both green and
yellow: a bell-shaped spectral reflectance in the central part
of the visible spectrum (typical of green) is mixed with a flat,
high spectral reflectance for long wavelengths (yellow).
“Peach” is another oil on wood by Matteo Massagrande.
It has been acquired under the same illumination conditions
and using the same setup as for “Tulips”. Its (R, G, B)
reconstruction can be seen in Figure 10.
Figure 11(b) shows the spectral reflectance of four points
(Figure 11(a)) chosen on this painting. The sky point (Point
1) shows the behavior typical of blue (a bell-shaped bump in
EURASIP Journal on Image and Video Processing
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
11
450 500 550 600 650 700 750 800
λ
(a)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
(b)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
(c)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
(d)
450 500 550 600 650 700 750 800
λ
(e)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
(f)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450 500 550 600 650 700 750 800
λ
Iod. + inc.
Iod. + hal.
Ref.
Iod. + inc.
Iod. + hal.
Ref.
(g)
(h)
Figure 7: Tabulated and reconstructed spectra using the metallic iodide+halogen and metallic iodide+incandescence illumination setups
for the (a) red, (b) orange, (c) yellow, (d) green, (e) cyan, (f) blue, (g) violet, and (h) purple tiles.
the 450–550 nm interval), but also reveals the presence of an
orange-red component. Point 2 is a typical pink, with light
reflected along the whole spectrum, but especially around
600 nm and in the orange-red region. Point 3 is a very dark
green, while Point 4 exhibits the characteristics of a dark
brown, with a behavior similar to that of orange-red but with
a smoother transition from low to higher reflectance values.
The last painting we have measured is “Bullfighter”, an oil
on canvas by another contemporary Italian painter, Vittorio
Buzzanca. The main difference between this painting and
the other two lies in the presence of a protective glass over
the canvas. The glass introduces effects of reflection and
dispersion, which can lead to a blurred reconstruction. The
acquisition configuration has thus been slightly modified in
order to obtain better results. The illumination sources have
been positioned at a more grazing angle with respect to the
painting’s surface, and the painting itself has been brought
closer to the camera. The (R, G, B) reconstruction obtained
with this setup is shown in Figure 12.
Figure 13(b) shows the spectral reflectance of the points
indicated in Figure 13(a). In this case, it is very easy to
recognize the spectral reflectance of a lighter and darker blue
12
EURASIP Journal on Image and Video Processing
1
2
3
4
(a)
Figure 8: “Tulips” by Matteo Massagrande (oil on wood).
1
(Points 1 and 4, resp.) and that of the two red spots (Points 2
and 3).
0.9
0.8
0.7
5.2. On-Field Validation. An on-field validation of our
measurement procedure has been performed by acquiring
a portion of a fresco painted by Girolamo di Romano,
known as “Romanino” in the Castello del Buonconsiglio in
Trento (Italy). Romanino was born in 1485 in Brescia (Italy),
a city in which one of the most important north Italian
schools of Renaissance painting had developed. During his
youth, he fell under the spell of Venetian painting, especially
that of Giorgione and Titian, and of Milanese painting.
As he matured, however, he developed a very personal
style, drawing inspiration especially from the very dramatic
pictorial style of German art, as demonstrated in the
magnificent cycle of paintings for Cremona cathedral (1519).
Romanino was a versatile artist; he painted on panels and
on canvas, but he favored the technique of fresco, the means
of expression which he found most congenial. In addition,
Vasari counted him among the most capable draughtsmen
of his time. In 1531 he was offered to decorate the Castello
del Buonconsiglio for the Prince Bishop of Trento, Bernard
Cles, and completed a large cycle of paintings, with secular
themes, in the castle.
The acquired scene, an (R, G, B) rendition of which is
shown in Figure 14, is a portion of a fresco representing a rest
after the hunt, with a servant (on the right) leaning towards
his master, holding a dove on his arm. We will call this
acquired portion “The dove”. The fresco is situated in a vault
of the “Volto sotto la Loggia”, one of the rooms of Magno
Palazzo, and its dimensions are approximately 2 m × 2 m.
The acquired area is positioned at a height of approximately
2.5 m. A picture of the acquisition setup can be seen in
Figure 15. It can be noted that the illumination sources are
placed several meters far from the target surface, also because
of the presence of the staircase. Therefore, two metallic iodide
lamps and two halogen lamps have been used to illuminate
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450
500
550
600
λ
Point 1
Point 2
650
700
750
800
Point 3
Point 4
(b)
Figure 9: (b) measured spectral reflectance of the points indicated
in (a) for the painting “Tulips”.
the scene. Figure 15 shows the two metallic iodide lamps
symmetrically mounted on a tripod.
As previously mentioned in Section 2, the acquisition
of the white signal is often difficult when acquiring large
scenes, and particularly difficult when acquiring a vault. As
it often happens, in this case we did not possess a sufficiently
large white panel to cover the whole scanned area. We have
therefore acquired the white signal in two different steps,
placing the white panel (mounted on a tripod) in different,
but partially overlapping, positions. The final reference
white has been obtained by exploiting a property of our
illumination system, that is, that the spectral content of the
illumination function (considered as the juxtaposition of the
EURASIP Journal on Image and Video Processing
13
1
2
3
4
(a)
Figure 10: “Peach” by Matteo Massagrande (oil on wood).
1
spectra of the two lamps) is constant [50]. This allows to
express the illumination function as
I x, y, λ = k x, y I0 (λ),
(14)
where (x, y) are the spatial coordinates, k(x, y) is a spatially
dependent coefficient accounting for the illumination intensity, and I0 (λ) is a known spectrally dependent function.
To recover the white signal, we have calculated k(x, y)
on the two panels, and fitted a plane in the least-square
sense to the two resulting surfaces. The interpolation of
the plane on the whole image support has been taken as
the reference white signal. Figure 16 shows the input and
output of the interpolation process. It is worth pointing out
that is clearly unrealistic to assume that of Figure 16(c) is
the actual illumination of the left part of the vault, which
lies on on a different, nonplanar surface. However, as no
paintings are present in that region, we have given up the
acquisition of the actual white signal for the presented study,
and used one single interpolated surface for the illumination
coefficients k. As the interpolated white signal satisfies the
physical constraints for the illumination on all its support,
the corresponding reflectance values are feasible, if not
correct.
Figure 17(b) shows the spectral reflectance of the points
shown in Figure 17(a). It can immediately be observed
that all the colors are quite dark, probably due to aging.
Nonetheless, the behavior of blue (Point 1), red (Point
2), yellow (Point 3), and green (Point 4) is still perfectly
recognizable.
6. Conclusions
Large-sized pictorial surfaces such as frescoed rooms or great
paintings represent a relevant aspect of artistic expression.
Their acquisition poses different challenges from those
encountered when acquiring regular paintings, the most
relevant dissimilarity being that in the former case the
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450
500
550
600
λ
Point 1
Point 2
650
700
750
800
Point 3
Point 4
(b)
Figure 11: (b) measured spectral reflectance of the points indicated
in (a) for the painting “Peach”.
illumination sources must be placed far from the acquired
scene. Therefore, the instruments and procedures used for
the multispectral acquisition of regular-sized paintings are
not suited to the acquisition of large surfaces. This paper
systematically addresses the issues involved in multispectral
acquisition of large-sized painted surfaces. Suitable acquisition instrumentation comprising an IS multispectral camera
mounted on a rotatory device is exemplified. As illumination
selection turns out to be a rather critical task, the issue is
discussed in detail. We have considered four illumination
setups: a halogen lamp alone (the illumination source
most commonly used for the multispectral acquisition of
paintings), an incandescence lamp alone, and two composite
14
EURASIP Journal on Image and Video Processing
Figure 14: “The dove” by Romanino.
Figure 12: “Bullfighter” by Vittorio Buzzanca (oil on canvas).
2
1
3
4
(a)
1
Figure 15: Acquisition setup in the Castello del Buonconsiglio,
Trento (Italy).
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450
500
550
600
λ
Point 1
Point 2
650
700
750
800
Point 3
Point 4
(b)
Figure 13: (b) measured spectral reflectance of the points indicated
in (a) for the painting “Bullfighter”.
setups comprising a metallic iodide lamp for the measurement of reflectance between 420 and 500 nm, and the
halogen or incandescence lamp for recovering the reflectance
between 500 and 800 nm. These illumination configurations
have been tested on the acquisition of the spectral reflectance
of a set of calibrated colored tiles. We have also examined
the delicate issue of performance evaluation and defined
the error as a function of wavelength, using a metrological
procedure to infer the uncertainty of the computed error
from the statistics of the measured variables. To quantify
the performance of reflectance measurements, we have used
the average error (AE) and the average error standard
deviation (AESTD), calculated for each illumination setup
and averaged over the eight-tile set. Although both the
combined illumination setups yield a good performance, the
best results have been obtained using the metallic iodide
lamp in combination with the incandescence lamp (|AE| <
0.02, AESTD ∼ 0.01). However, considerations of practical
EURASIP Journal on Image and Video Processing
15
×104
2.2
2.1
200
2
400
1.9
600
1.8
800
1.7
1.6
1000
100
200
300
400
500
(a)
×104
2.2
2.1
200
2
400
1.9
600
1.8
800
1.7
1.6
1000
100
200
300
400
500
(b)
×104
2.2
100
2.1
200
300
2
400
1.9
500
600
1.8
700
800
1.7
900
1.6
1000
100
200
300
400
500
(c)
Figure 16: (a) k coefficient corresponding to the white panel in position 1; (b) k coefficient corresponding to the white panel in position 2;
(c) interpolated planar surface for the coefficients k(p).
nature suggest that in on-field applications, the employment
of halogen lamps is preferable to that of incandescence
bulbs. Both illumination configurations have therefore been
validated in laboratory and in-field conditions. The combi-
nation of metallic iodide and incandescence lamps has been
deployed for the acquisition of three paintings from Italian
contemporary artists in a controlled environment. The setup
featuring the metallic iodide and halogen lamps has then
16
EURASIP Journal on Image and Video Processing
1
2
4
3
(a)
1
0.9
0.8
0.7
0.6
R 0.5
0.4
0.3
0.2
0.1
0
450
500
550
600
λ
Point 1
Point 2
650
700
750
800
Point 3
Point 4
(b)
Figure 17: (b) measured spectral reflectance of the points indicated
in (a) for the fresco “The dove”.
been used for the on-field acquisition of a large portion of
a fresco of the Castello del Buonconsiglio in Trento (Italy).
Acknowledgments
The authors would like to thank Dr. Fabio Remondino for
letting them perform the acquisitions in the Castello del
Buonconsiglio. They are also grateful to Carlo Tona and
Matteo Caldon for their contribution to the multispectral
measurements.
References
[1] M. Barni, A. Pelagotti, and A. Piva, “Image processing for
the analysis and conservation of paintings: opportunities and
challenges,” Signal Processing Magazine, vol. 22, no. 5, pp. 141–
144, 2005.
[2] B. Hill, “Multispectral color technology: a way toward highdefinition color image scanning and encoding,” in Proceedings
of the 6th International Conference on Computer Vision (ICCV
’98), vol. 3409 of Proceedings of SPIE, pp. 2–13, Mumbai, India,
January 1998.
[3] G. Novati, P. Pellegri, and R. Schettini, “An affordable multispectral imaging system for the digital museum,” International
Journal on Digital Libraries, vol. 5, no. 3, pp. 167–178, 2005.
[4] A. Pelagotti, A. D. Mastio, A. D. Rosa, and A. Piva, “Multispectral imaging of paintings,” Signal Processing Magazine, vol. 25,
pp. 25–36, 2008.
[5] K. Martinez, J. Cupitt, D. Saunders, and R. Pillay, “Ten years
of art imaging research,” Proceedings of the IEEE, vol. 90, no. 1,
pp. 28–41, 2002.
[6] K. Martinez and A. Hamber, “Towards a colorimetric digital
image archive for the visual arts,” in Proceedings of the IEEE,
vol. 1073, pp. 114–121, January 1989.
[7] K. Martinez, “High resolution digital imaging of paintings.
The Vasari project,” Microcomputers for Information Management, vol. 8, no. 4, pp. 277–283, 1991.
[8] K. Martinez, J. Cupitt, and D. Saunders, “High resolution
colorimetric imaging of paintings,” in Proceedings of the
Meeting on Cameras, Scanners, and Image Acquisition Systems,
vol. 1901 of Proceedings of SPIE, pp. 25–36, San Jose, Calif,
USA, February 1993.
[9] J. Farrell, J. Cupitt, D. Saunders, and B. Wandell, “Estimating
spectral reflectances of digital artwork,” in Proceedings of the
International Symposium on Multispectral Imaging and Color
Reproduction for Digital Archives, pp. 58–64, Chiba, Japan,
October 1999.
[10] J. Cupitt, K. Martinez, and D. Saunders, “Methodology for art
reproduction in colour: the MARC project,” Computers and
the History of Art Journal, vol. 6, no. 2, pp. 1–20, 1996.
[11] F. Schmitt, H. Brettel, and J. Hardeberg, “Multispectral
imaging development at ENST,” Display and Imaging, vol. 8,
pp. 261–268, 2000.
[12] H. Maitre, F. J. Schmitt, and J. P. Crettez, “High quality
imaging in museum: from theory to practice,” in Proceedings of
the Conference of the Very High Resolution and Quality Imaging,
vol. 3025, pp. 30–39, San Jose, Calif, USA, February 1997.
[13] H. Maı̂tre, F. Schmitt, and C. Lahanier, “15 years of image
processing and the fine arts,” in Proceedings of the IEEE
International Conference on Image Processing (ICIP ’01), vol.
1, pp. 557–561, Thessaloniki, Greece, October 2001.
[14] F. Schmitt, “High quality digital color images,” in Proceedings
of the 5th International Conference on High Technology (CHIBA
’96), pp. 55–62, Chiba, Japan, September 1996.
[15] H. Maı̂tre, F. Schmitt, J. Crettez, Y. Wu, and J. Y. Hardeberg,
“Spectrophotometric image analysis of fine art paintings,” in
Proceedings of the 4th IS&T—SID Colour Imaging Conference,
pp. 50–53, Scottsdale, Ariz, USA, November 1996.
[16] J. Hardeberg, F. Schmitt, H. Brettel, J.-P. Crettez, and H.
Maı̂tre, “Multispectral image acquisition and simulation of
illuminant changes,” Colour Imaging: Vision and Technology,
pp. 145–164, 1999.
[17] J. Hardeberg, F. Schmitt, and H. Brettel, “Multispectral image
capture using a tunable filter,” in Proceedings of the 4th
Conference of Color Imaging: Device Independent Color, Color
Hardcopy and Graphic Arts, vol. 3963, pp. 77–88, July 2000.
[18] J. Hardeberg, “Multispectral color image acquisition,” in
Proceedings of the Norwegian Signal Processing Symposium
(NORSIG ’01), pp. 77–82, Trondheim, Norway, October 2001.
EURASIP Journal on Image and Video Processing
[19] J. Hardeberg, Acquisition and reproduction of color images—
colorimetric and multispectral approaches, Ph.D. dissertation,
Universal Publishers, Parkland, Fla, USA, 2001.
[20] Y. Yokoyama, T. N. H. Haneishi, Y. Miyake, J. Hayashi, and
M. Saito, “A new color management system based on human
perception and its application to recording and reproduction
of art paintings,” in Proceedings of the 5th Color Imaging
Conference: Color Science, Systems and Applications, pp. 169–
172, Scottsdale, Ariz, USA, November 1997.
[21] H. Haneishi, T. Hasegawa, A. Hosoi, Y. Yokoyama, N.
Tsumura, and Y. Miyake, “System design for accurately
estimating the spectral reflectance of art paintings,” Applied
Optics, vol. 39, no. 35, pp. 6621–6632, 2000.
[22] H. Haneishi, T. Hasegawa, N. Tsumura, and Y. Miyake,
“Design of color filters for recording artworks,” in Proceedings
of the IS&T 50th Annual Conference, pp. 369–372, Springfield,
Va, USA, 1997.
[23] J. P. Mohen, M. Menu, and B. Mottin, Mona Lisa: Inside the
Painting, Harry N. Abrams, New York, NY, USA, 2006.
[24] A. Ribes, F. Schmitt, R. Pillay, and C. Lahanier, “Calibration
and spectral reconstruction for CRISATEL: an art painting
multispectral acquisition system,” Journal of Imaging Science
and Technology, vol. 49, no. 6, pp. 563–573, 2005.
[25] G. S. E. Ciliberto, Ed., Modern Analytical Methods in Art and
Archeology, John Wiley & Sons, New York, NY, USA, 2000.
[26] S. Tominaga, “Spectral imaging by a multichannel camera,”
Journal of Electronic Imaging, vol. 8, no. 4, pp. 332–341, 1999.
[27] F. H. Imai, M. R. Rosen, and R. S. Berns, “Multi-spectral
imaging of van Gogh’s selfportrait at the National Gallery
of Art, Washington, D.C.,” in Proceedings of the IS& PICS
Conference, pp. 185–189, Springfield, Va, USA, October 2001.
[28] H. Sugiura, T. Kuno, N. Watanabe, N. Matoba, J. Hayashi, and
Y. Miyake, “Development of a multispectral camera system,”
in Proceedings of the Conference of Sensors and Camera Systems
for Scientific, Industrial, and Digital Photography Applications,
M. M. Blouke, N. Sampat, G. M. Williams, and T. Yeh, Eds.,
vol. 3965 of Proceedings of SPIE, pp. 331–339, May 2000.
[29] The archimedes palimpsest project, 2008, http://www.archimedespalimpsest.org/index.html.
[30] G. Antonioli, F. Fermi, C. Oleari, and R. Riverberi, “Spectrophotometric scanner for imaging of paintings and other
works of art,” in Proceedings of the 2nd European Conference on
Colour in Graphics, Imaging an Vision, pp. 219–224, Aachen,
Germany, April 2004.
[31] P. Carcagni, A. D. Patria, R. Fontana, et al., “Multispectral
imaging of paintings by optical scanning,” Optics and Lasers
in Engineering, vol. 45, no. 3, pp. 360–367, 2007.
[32] N. Brusco, S. Capeleto, M. Fedel, et al., “A system for
3D modeling frescoed historical buildings with multispectral
texture information,” Machine Vision and Applications, vol. 17,
no. 6, pp. 373–393, 2006.
[33] P. D. Burns and R. S. Berns, “Analysis multispectral image
capture,” in Proceedings of the 4th IS&T/SID Color Imaging
Conference, pp. 19–22, Scottsdale, Ariz, USA, 1995.
[34] “Guide to the expression of uncertainty in measurement,” ISO
Std, 17.020, 2007.
[35] “Specim imaging spectrograph brochure,” 2007, http://www
.specim.fi/products/spetral-imaging-products/imaging-spectrographs.html.
[36] W. L. Wolfe, Introduction to Imaging Spectrometers, vol. 25 of
SPIE Tutorial Text, Bellingham, Wash, USA, 1997.
17
[37] F. E. Nicodemus, J. C. Richmond, J. Hsia, I. W. Ginsberg, and
T. Limperis, “Geometrical considerations and nomenclature
for reflectance,” Tech. Rep., National Bureau of Standards,
Inst. for Basic Standards, Washington, DC, USA, 1977.
[38] G. C. Holst and T. L. Lomheim, CMOS/CCD Sensors and
Camera Systems, JCD Publishing/SPIE Press, Bellingham,
Wash, USA, 2007.
[39] M. Caldon, M. Fedel, L. Poletto, N. Brusco, G. Cortelazzo, and
A. Paviotti, “Simultaneous acquisition of 3D shape and color
texture from large frescoed areas,” in Proceedings of the O3A:
Optics for Arts, Architecture, and Archaeology Conference, vol.
6618 of Proceedings of SPIE, pp. J.1–J.10, Munich, Germany,
June 2007.
[40] E. C. Carter and F. W. Billmeyer, “Material standards and their
use in color measurements,” Color Research and Application,
vol. 4, pp. 96–100, 1979.
[41] R. S. Berns and K. H. Petersen, “Empirical modeling of systematic spectrophometric errors,” Color Research and Application,
vol. 13, no. 4, pp. 243–256, 1988.
[42] I. Nimeroff, “Propagation of errors in spectrophometric
colorimetry,” Journal of the Optical Society of America, vol. 43,
no. 6, pp. 531–533, 1953.
[43] M. D. Fairchild and L. Reniff, “Propagation of random
errors in spectrophometric colorimetry,” Color Research and
Application, vol. 16, no. 6, pp. 360–367, 1991.
[44] P. D. Burns and R. S. Berns, “Error propagation analysis
in color measurement and imaging,” Color Research and
Application, vol. 22, no. 4, pp. 280–289, 1997.
[45] M. J. Vrhel and H. J. Trussell, “Optimal color filters in the
presence of noise,” IEEE Transactions on Image Processing, vol.
4, no. 6, pp. 814–823, 1995.
[46] “Labsphere spectralon color standards,” 2006, http://www
.labsphere.com/productdetail.aspx?id=897.
[47] J. Brauers, N. Schulte, A. A. Bell, and T. Aach, “Multispectral
high dynamic range imaging,” in Proceedings of the IS&T/SPIE
Electronic Imaging, pp. 680704-1–680704-12, San Jose, Calif,
USA, January 2008.
[48] N. Gat, “Imaging spectroscopy using tunable filters: a review,”
in Proceedings of the 7th Wavelet Applications Conference, vol.
4056 of Proceedings of SPIE, pp. 50–64, Orlando, Fla, USA,
April 2000.
[49] G. Wyszecki and W. S. Stiles, Color Science—Concepts and
Methods, Quantitative Data and Formulae, John Wiley & Sons,
New York, NY, USA, 1982.
[50] A. Paviotti, Acquisition and processing of multispectral data for
texturing 3D models, Ph.D. dissertation, University of Padova,
Padova, Italy, March 2009.